WHEN TO
RESET?

A small, rule-based experiment, not a trained predictor. We start with an assumed 12% daily baseline, then change an hourly event rate using the strongest signal in each category.

Weights multiply the hourly rate exponentially. Related incidents add at most +0.55; launches add +0.20. Repeated posts do not stack within a category. Ordinary signals halve in 24 hours; undated promises halve in 12. Dated promises only affect their target day in Pacific time. An observed reset retires earlier hints and promises.

Banked reset rule: a first-party announcement that cards are being loaded or granted counts as an observed reset event today. A future plan adds +2.20 to the hourly model and decays in 12 hours; it never confirms delivery. 100% confirms the public announcement, not delivery to your account.

Today means the time left before midnight in your selected timezone. Tomorrow is the following calendar day. Predictions are capped at 95%; 100% is reserved for a first-party report of an automatic reset or a banked reset grant. The weights and baseline are design choices, not measured historical frequencies. We have no backtested accuracy score. Read the original before relying on a classification.

We track extra Codex reset events, including banked reset grants and automatic shared resets. A banked reset is saved for later use and does not immediately refill your quota. Scheduled personal refills are separate. This page cannot see your account or reset your limits.